Collision Avoidance of Multiple Moving Agents by Adapting the A* Algorithm

The A* algorithm is one of the most popular pathfinding algorithms. The basic algorithm can reliably find the best path for an agent in a static environment. However, there is only limited knowledge on how the algorithm behaves in a dynamic context. One important dynamic element in a pathfinding problem might be other agents moving simultaneously in the same environment, such as in the application scenarios of various real-time strategy games. With the basic
$\mathbf{A}^{*}$
algorithm, these agents could collide, especially when moving around an obstacle or through a narrow passage. To avoid collisions, a modified agent control is proposed. This extension consists of introducing a waiting time when an agent moves to a place where another agent is already located. The waiting time operator is directly included into the optimization algorithm, stimulating the algorithm to search for alternative routes that avoid these waiting times. The resulting routes might be longer in distance but could be faster because the agent avoids the other blockade. Experiments with different settings indicate that the algorithm achieves this goal: In all settings, including narrow passages, collisions between agents were no longer detected. Furthermore, searching for alternative routes helps the algorithm find paths which are more than 10% faster. The duration of the slowest path can also be reduced in 80% of cases.